Database release deployment on OSG

Database Deployment on OSG
Yuri Smirnov
BNL
US ATLAS DDM operations and MC production Workshop, BNL
September 28-29, 2006
Outline
 ATLAS MySQL services for OSG at US ATLAS Tier1
 Panda Database services on OSG
 Database replication CERN -> BNL
 Monitoring tools for MySQL servers: MySQLStat
 Database release deployment on OSG
 Production: 11.0.X and database deployment on OSG
 Production: 12.0.X (project-built) and database deployment
 New DB deployment approach in 12.0.3
Yuri Smirnov (BNL)
September 28,2006
2
BNL MySQL DB services for US ATLAS
and OSG
 BNL Tier1 not only contributes to ATLAS OSG production (half of U.S.
production in 2006 was done at Tier1, remainder at Tier2 sites), but
provides many different reliable services, including MySQL DataBases
support
 ATLAS production and development MySQL servers for OSG at BNL:
 dbpro production cluster (PandaDB)
 adbpro development cluster (PandaDevDB, grid-enable MySQL development)
 db1 (ConditionsDB IOV+payload, DBs for subdetectors LAr, TileCal, etc.,
LRC, DataSetDB, DialDB, Geometry DB Nova,etc.)
 dbdevel1 (ATLASDD, MagdaFC)
 dbdevel2 (TAG DB, PandaArchiveDB, PandaLogDB)
 dms01/dms02 ( DQ2 services and catalogs, will support Production LRC soon)
 gums/vo (GUMS and VOMS db for OSG and LCG)
Yuri Smirnov (BNL)
September 28,2006
3
Panda database production service
 dbpro MySQL cluster: 2 nodes dbpro01 and dbpro02
 easy recovery: if one node goes down, the 2-nd node contunues to
work without interruption in providing the database service

in-memory technology for Panda DB (NDB engine)
 to provide quick reliable access to data PandaDB keeps information
about production and analysis jobs completed withing ~ one week.
 PandaArchiveDB on dbdevel2 server:

Uses the same structure as in-memory PandaDB,

Disk-resident tables (MyISAM engine)

Keeps track of all production and user analysis jobs completed/failed on OSG since
the beginnig 2006

Size is ~1GB at the moment
 PandaLogDB on dbdevel2 stores log-file extracts of failed jobs.
 Excellent monitoring tool Panda-monitor to both PandaDB and
PandaArchiveDB developed by Torre Wenaus
Yuri Smirnov (BNL)
September 28,2006
4
Panda-monitor
Yuri Smirnov (BNL)
September 28,2006
5
Database replication at BNL

Oracle – Oracle stream-replication (presentation of Iris)

MySQL – MySQL replication:


DataBases: Geometry DB, Conditions DB

collected the first experience CERN-BNL ATLAS DB replication

procedure using both mysqldump and on-line replication

thanks to Sasha Vaniachine and Vakho Tsulaia for participation
Oracle – MySQL replication:

DataBase: TAG DB.

use case : Oracle CERN to MySQL BNL (push)

tool: Octopus replicator ( Java-based extraction, transformation and loading)

thanks to Julius Hrivnac, Ed May and Kristo Karr for successful collaboration

More details in Twiki: https://uimon.cern.ch/twiki/bin/view/Atlas/DatabaseReplication
Yuri Smirnov (BNL)
September 28,2006
6
MySQL servers monitoring
We use different
monitoring tools for
MySQL servers and
services:
-MySQLStat (status of
different BNL, CERN,
OSG, LCG servers )
-Nagios
-Ganglia
Yuri Smirnov (BNL)
September 28,2006
7
MySQL servers monitoring
Yuri Smirnov (BNL)
September 28,2006
8
Database release deployment on OSG
 CSC Panda production on OSG uses ATLAS software
releases installation (dist.kit) at Tier1 and Tier2s
 Two types of production releases:

11-branch: 11.0.42, 11.0.5; txt jobOptions, shell-based trfs;

12-branch: 12.0.1, 12.0.2, 12.0.3 (soon): project-built releases,
python-based trfs and jobOptions.
 Production jobs work with 2 ATLAS core DataBases:

Geometry Database;

Conditions Database (both IoV and payload)
Yuri Smirnov (BNL)
September 28,2006
9
11.0.X DB releases deployment
 11.0.X release itself and job-transformations use two different
areas:

$APP/atlas_app/atlas_rel/11.0.X/

$APP/atlas_app/atlas_rel/kitval/KitValidation/JobTransforms/JobTransforms
-11-00-0X-Y
 Geometry Database (in SQLite-files) is a part of ATLAS-release
installation (dist. kit), for instance:

in 11.0.5 geomDB_sqlite file, size ~11MB

Location: $APP/atlas_app/atlas_rel/11.0.5/atlas/offline/data
 Conditions Database runs on MySQL sever:

Master at CERN: atlasdbdev.cern.ch

Replica at BNL: db1.usatlas.bnl.gov
Yuri Smirnov (BNL)
September 28,2006
10
12.0.X DB releases deployment
 12.0.X releases use job-transformations as a part of the installation
(and new versions can be easily added):

$APP/atlas_app/atlas_rel/12.0.X/

$APP/atlas_app/atlas_rel/12.0.X/AtlasProduction/12.0.X.Y
 Geometry DB (Sqlite-files) and Conditions DB come with the
installation as a Database Release 2.Z (default version):

$APP/atlas_app/atlas_rel/12.0.X/DBRelease/2.Z

DB release cycle may be not synchronized with ATLAS release (one ATLAS release
can work with several DB releases)
 If a new version of DB release is available it can be installed on OSG
as a new subdirectory under the same ATLAS-release, for instance:

$APP/atlas_app/atlas_rel/12.0.2/DBRelease/2.6

$APP/atlas_app/atlas_rel/12.0.2/DBRelease/2.7
Yuri Smirnov (BNL)
September 28,2006
11
New DB deployment approach in
12.0.3
 New technology for production DB deployment in
release 12.0.3 is prepared by Martin and Vakho.
 Can be tested already now with 12.0.2 under
12.0.2.3 transformations (DB-release 2.7.1).
 Idea of dynamic DB-release installation:

trf itself (not the installation script) should check if the
corresponding Database Release (SQLite files) is available and
install it (through DQ2 subscriptions) directly on the WN where the
job using this trf runs. Can be removed later when necessary.
Yuri Smirnov (BNL)
September 28,2006
12